1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -S -mtriple=x86_64-unknown -mattr=+avx512vl -passes=slp-vectorizer -slp-threshold=-3 | FileCheck %s 3 4 5define i32 @test(ptr noalias %in, ptr noalias %inn, ptr %out) { 6; CHECK-LABEL: @test( 7; CHECK-NEXT: [[TMP7:%.*]] = load <4 x i8>, ptr [[IN:%.*]], align 1 8; CHECK-NEXT: [[TMP11:%.*]] = load <4 x i8>, ptr [[INN:%.*]], align 1 9; CHECK-NEXT: [[TMP8:%.*]] = sext <4 x i8> [[TMP7]] to <4 x i16> 10; CHECK-NEXT: [[TMP12:%.*]] = sext <4 x i8> [[TMP11]] to <4 x i16> 11; CHECK-NEXT: [[TMP13:%.*]] = sub <4 x i16> [[TMP12]], [[TMP8]] 12; CHECK-NEXT: [[TMP15:%.*]] = call <4 x i16> @llvm.abs.v4i16(<4 x i16> [[TMP13]], i1 false) 13; CHECK-NEXT: store <4 x i16> [[TMP15]], ptr [[OUT:%.*]], align 2 14; CHECK-NEXT: ret i32 undef 15; 16 %load.1 = load i8, ptr %in, align 1 17 %gep.1 = getelementptr inbounds i8, ptr %in, i64 1 18 %load.2 = load i8, ptr %gep.1, align 1 19 %gep.2 = getelementptr inbounds i8, ptr %in, i64 2 20 %load.3 = load i8, ptr %gep.2, align 1 21 %gep.3 = getelementptr inbounds i8, ptr %in, i64 3 22 %load.4 = load i8, ptr %gep.3, align 1 23 %load.5 = load i8, ptr %inn, align 1 24 %gep.4 = getelementptr inbounds i8, ptr %inn, i64 1 25 %load.6 = load i8, ptr %gep.4, align 1 26 %gep.5 = getelementptr inbounds i8, ptr %inn, i64 2 27 %load.7 = load i8, ptr %gep.5, align 1 28 %gep.6 = getelementptr inbounds i8, ptr %inn, i64 3 29 %load.8 = load i8, ptr %gep.6, align 1 30 %sext1 = sext i8 %load.1 to i32 31 %sext2 = sext i8 %load.2 to i32 32 %sext3 = sext i8 %load.3 to i32 33 %sext4 = sext i8 %load.4 to i32 34 %sext5 = sext i8 %load.5 to i32 35 %sext6 = sext i8 %load.6 to i32 36 %sext7 = sext i8 %load.7 to i32 37 %sext8 = sext i8 %load.8 to i32 38 %sub1 = sub i32 %sext1, %sext5 39 %sub2 = sub i32 %sext2, %sext6 40 %sub3 = sub i32 %sext7, %sext3 41 %sub4 = sub i32 %sext8, %sext4 42 %call1 = call i32 @llvm.abs(i32 %sub1, i1 true) 43 %call2 = call i32 @llvm.abs(i32 %sub2, i1 true) 44 %call3 = call i32 @llvm.abs(i32 %sub3, i1 true) 45 %call4 = call i32 @llvm.abs(i32 %sub4, i1 true) 46 %t1 = trunc i32 %call1 to i16 47 %t2 = trunc i32 %call2 to i16 48 %t3 = trunc i32 %call3 to i16 49 %t4 = trunc i32 %call4 to i16 50 %gep.8 = getelementptr inbounds i16, ptr %out, i64 1 51 %gep.9 = getelementptr inbounds i16, ptr %out, i64 2 52 %gep.10 = getelementptr inbounds i16, ptr %out, i64 3 53 store i16 %t1, ptr %out, align 2 54 store i16 %t2, ptr %gep.8, align 2 55 store i16 %t3, ptr %gep.9, align 2 56 store i16 %t4, ptr %gep.10, align 2 57 58 ret i32 undef 59} 60